“2019-05-24”版本间的差异

来自cslt Wiki
跳转至： 导航、 搜索
 
（6位用户的8个中间修订版本未显示）
第9行： 第9行：
 
|Yibo Liu
 
|Yibo Liu
 
||  
 
||  
−
*
+
* Released a version on github https://github.com/CSLT-THU/Vivi_3.0 . Added Readme.
 
||
 
||
−
*  
+
* Do some experiments to solve the problem of bad effect on the last 2 sentences, e.g. modify training method, modify loss function.
 
||
 
||
 
*   
 
*   
第36行： 第36行：
 
|Jiayao Wu
 
|Jiayao Wu
 
||  
 
||  
−
*  
+
* made further research on sentence-level resample boosting and trying frame-level method.
 
||
 
||
−
*  
+
* prepare for the weekly report.
 +
*keep on doing frame-level research.
 
||  
 
||  
 
*  
 
*  
第49行： 第50行：
 
|Zhaodi Qi
 
|Zhaodi Qi
 
||  
 
||  
−
*  
+
* Do experiment on Multi-language BN
 
||  
 
||  
−
*  
+
* Summary of the experiment
 
||
 
||
 
*   
 
*   
第62行： 第63行：
 
|Jiawei Yu
 
|Jiawei Yu
 
||  
 
||  
−
* Test part of ivector and dvector in-set LID experiment.
+
* Done part of ivector and dvector in-set LID experiment.
−
* Experiment different test utterance lengths(1s, 3s, full length) for zero-resource LID.
+
* Experimented different test utterance lengths(1s, 3s, full length) for zero-resource LID.
 
||  
 
||  
 
* Test the in-set language for different test utterance length(1s, 3s).
 
* Test the in-set language for different test utterance length(1s, 3s).
第89行： 第90行：
 
|Xueyi Wang
 
|Xueyi Wang
 
||
 
||
−
*  
+
* reproducing the experiments of v-vector and c-vector; result is close to the one in the paper
 
||  
 
||  
−
*  
+
* get down to unsupervised PLDA
 
||
 
||
 
*   
 
*   
第102行： 第103行：
 
|Ziya Zhou
 
|Ziya Zhou
 
||  
 
||  
−
*
+
* Downloaded videos of 258 celebrities
 +
* Complete the downloading task by next Monday
 
||
 
||
−
*  
+
* Edit the videos
 
||
 
||
 
*   
 
*   
第114行： 第116行：
 
|kaicheng li
 
|kaicheng li
 
||
 
||
−
*  
+
* Collect celebrity audio data （100）
 
||  
 
||  
 +
* Continue collecting celebrity audio data
 +
||
 
*  
 
*  
−
||
 
−
* 
 
 
|-
 
|-
  

2019年5月24日 (五) 00:15的最后版本

People This Week Next Week Task Tracking (DeadLine)
Yibo Liu
  • Do some experiments to solve the problem of bad effect on the last 2 sentences, e.g. modify training method, modify loss function.
Xiuqi Jiang
  • Compiling the data of planning and training set, unifying data formats
  • Re-plan training set using planning results
  • Doing summary work
Jiayao Wu
  • made further research on sentence-level resample boosting and trying frame-level method.
  • prepare for the weekly report.
  • keep on doing frame-level research.
Zhaodi Qi
  • Do experiment on Multi-language BN
  • Summary of the experiment
Jiawei Yu
  • Done part of ivector and dvector in-set LID experiment.
  • Experimented different test utterance lengths(1s, 3s, full length) for zero-resource LID.
  • Test the in-set language for different test utterance length(1s, 3s).
Wenwei Dong
  • -
  • Try speaker level correlation.
  • Collect different language background speaker's English data to train info-GAN.
Xueyi Wang
  • reproducing the experiments of v-vector and c-vector; result is close to the one in the paper
  • get down to unsupervised PLDA
Ziya Zhou
  • Downloaded videos of 258 celebrities
  • Complete the downloading task by next Monday
  • Edit the videos
kaicheng li
  • Collect celebrity audio data （100）
  • Continue collecting celebrity audio data
Haolin Chen
  • Maths: statistical inference, integer programming
  • Continue learning statistics
Haoran Sun
  • Completed the testing table of asr engine
  • Test the concurrence of asr engine
取自“http://index.cslt.org/mediawiki/index.php?title=2019-05-24&oldid=33026”